Merge branch 'mesa_7_5_branch'

[mesa.git] / src / gallium / auxiliary / rtasm / rtasm_x86sse.c

diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
index 3bba9dcc076c27529bfbc142acabb87fd8680aae..1acf3c373ebee1700b74463c834d1363a2098e3d 100644 (file)
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -26,7 +26,7 @@
 #if defined(PIPE_ARCH_X86)
 
 #include "pipe/p_compiler.h"
-#include "pipe/p_debug.h"
+#include "util/u_debug.h"
 #include "util/u_pointer.h"
 
 #include "rtasm_execmem.h"
@@ -371,7 +371,11 @@ void x86_jcc( struct x86_function *p,
    DUMP_I(cc);
    
    if (offset < 0) {
-      assert(p->csr - p->store > -offset);
+      /*assert(p->csr - p->store > -offset);*/
+      if (p->csr - p->store <= -offset) {
+         /* probably out of memory (using the error_overflow buffer) */
+         return;
+      }
    }
 
    if (offset <= 127 && offset >= -128) {
@@ -675,6 +679,44 @@ void x86_and( struct x86_function *p,
  * SSE instructions
  */
 
+void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 0, ptr);
+}
+
+void sse_prefetch0( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 1, ptr);
+}
+
+void sse_prefetch1( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 2, ptr);
+}
+
+void sse_movntps( struct x86_function *p, 
+                  struct x86_reg dst,
+                  struct x86_reg src)
+{
+   DUMP_RR( dst, src );
+
+   assert(dst.mod != mod_REG);
+   assert(src.mod == mod_REG);
+   emit_2ub(p, 0x0f, 0x2b);
+   emit_modrm(p, src, dst);
+}
+
+
+
 
 void sse_movss( struct x86_function *p,
                struct x86_reg dst,
@@ -951,6 +993,15 @@ void sse_pmovmskb( struct x86_function *p,
    emit_modrm(p, dst, src);
 }
 
+void sse_movmskps( struct x86_function *p,
+                   struct x86_reg dst,
+                   struct x86_reg src)
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x50);
+   emit_modrm(p, dst, src);
+}
+
 /***********************************************************************
  * SSE2 instructions
  */